Contrastive Learning for Seismic Horizon Tracking with Domain-Specific Priors

2026-06-15 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Environmental Science & Earth Systems, Geophysics · Depth: Expert, quick

Summary

A new self-supervised method addresses limitations in 3D seismic horizon tracking by fusing signal-based propagators with texture-driven deep learning models. This approach leverages signal-derived local horizon correspondences as domain-specific priors to train a deep model, overcoming the typical need for extensive labeled data. The method estimates reliable trace-to-trace flows from reflector slopes, forming positive pairs for a contrastive objective, with training focused on high-confidence areas, potentially using a fault mask. Its primary goal is to maintain horizon identity across discontinuities, rather than inferring ambiguous correspondences. This results in voxel-wise embeddings that ensure local signal continuity and enable horizon propagation through similarity search, even across faults. Evaluations on the public F3 dataset and a faulted synthetic dataset demonstrate lower mean absolute error (MAE) compared to unsupervised baselines and competitive results against a semi-supervised technique using only one labeled slice.

Key takeaway

For Geophysical Data Scientists or ML Engineers tasked with automating seismic horizon tracking, especially in faulted regions, you should consider integrating this self-supervised contrastive learning approach. It offers a robust method to propagate horizons beyond discontinuities without extensive labeled data, potentially reducing manual interpretation time and improving accuracy. Evaluate its performance on your specific datasets, utilizing signal-derived priors and fault masks to enhance model training.

Key insights

Fusing signal-based priors with contrastive learning enables robust seismic horizon tracking across faults.

Principles

• Signal-derived priors enhance texture-based deep models.
• Contrastive learning can preserve identity across discontinuities.
• Restricting training to high-confidence areas improves robustness.

Method

Estimate trace-to-trace flows from reflector slopes. Use these to form positive pairs in a contrastive objective. Train a deep model, restricting to high-confidence neighborhoods, optionally with a fault mask.

In practice

• Apply contrastive learning for fault-robust seismic tracking.
• Integrate reflector slope data as domain-specific priors.
• Use fault masks to refine training near discontinuities.

Topics

• Seismic Horizon Tracking
• Contrastive Learning
• Self-Supervised Learning
• Deep Learning
• Geophysical Data Analysis
• Fault Detection

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Physics-Guided Spatiotemporal Learning for Coastal Wave Peak Period Estimation from Video — Physics-guided deep spatiotemporal learning from video can accurately and consistently estimate coastal wave peak periods, overcoming traditional monitoring limitations.
• GNSS-FM: A Self-Supervised Foundation Model for Daily GNSS Displacement Time Series — GNSS-FM introduces self-supervised pretraining to overcome labeled data scarcity in GNSS time series analysis, outperforming supervised baselines.
• From Short Histories to Long Futures: Horizon-Aware Graph Neural Networks for Long Horizon Forecasting — A multi-horizon GNN emulator improves long-range geophysical forecasting by learning state transitions and predicting increments.
• WaveDINO: Learning-Based Atmospheric Correction of Unwrapped InSAR Interferograms Validated by GNSS: Results at Laguna del Maule and Campi Flegrei Volcanoes — WaveDINO employs hybrid training and DINOv3 features for superior learning-based atmospheric correction in InSAR.
• Domain-Guided Prompting of the Segment Anything Model for Seismic Interpretation: The Role of Attributes, Visualization, and Hybrid Prompts — Zero-shot SAM adaptation for seismic interpretation is achieved by combining domain-guided visualization with hybrid prompting.
• Advanced Flood Prediction with Physics-Guided Deep Learning: Combining UNet, FNO, and SAR/Optical Imagery — Embedding hydrodynamic principles into deep learning models significantly improves flood prediction accuracy and physical consistency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.